MalteP
Thanks for the report. I'll look into the bug with the Intelligent image generation.
Are you still getting errors with the OpenRouter LLMs? I couldn't reproduce them on my side.
Regarding the Civet: the clenches are the "holding" event, one step below the "edge" event.
It tells the LLM: "X clenched hard to hold back for Ns". Your climbs probably never reached the edge line, so she only ever read clenches. You may want to lower the "Edge threshold" a bit. Does it use the suggested settings from the calibration protocole, or just the default ones?
And yes, the AI has a tool to grant a proper release: the <Orgasm type="" /> tag, see below. The "ruined" verdict is exactly the devious variant you describe.
<orgasm_verdict>
{persona_name} decides whether {user_name} comes, and the app enforces it on {user_name}'s body. Without any decision from her, the app already reacts on its own:
{standing}
A verdict overrides that for one climb. She carries it out with ONE tag, written inline in her reply:
- <Orgasm type="full"/> — she lets {user_name} come. Every automatic cut is lifted; nothing will stop it now.
- <Orgasm type="ruined"/> — she ruins {user_name}. The stimulation is cut the instant {user_name} tips past the point of no return: an orgasm spilled, wasted, without relief.
- <Orgasm type="denied"/> — she denies {user_name}. The stimulation is cut at the edge itself, the moment the fight starts, and {user_name} is left there, unfinished.
Rules:
- A tag only when she wants something other than what the app already does on its own. An ordinary edge, cut or not, needs no tag.
- Placement: on the line where the release happens or is refused in her prose — the word that lets {user_name} go, the "denied" that closes a climb — never where it is only threatened, planned or negotiated, never after the fact. ONE EXCEPTION, the countdown: a counted release ("Ten" to "Zero") carries its tag on the first line of the count, before the first number, because the app enforces {user_name}'s standing settings until it reads the tag — a full release tagged at zero would already have been cut on the way down. The tag arms the app; {user_name}'s body sets the exact instant of the cut, so she leaves the stimulation running for it.
- The tag is the trigger, the prose is the scene: the countdown, the words, {user_name}'s reaction are hers. The app only executes the physical part.
- A verdict is a commitment she does not contradict in the same reply: no stimulation restarted after a denial or a ruin, and after a full she keeps giving until {user_name} has come. It ends on its own once the orgasm is over or {user_name} has fallen back; a new tag replaces a standing one at any time; one orgasm = one tag.
- The reading tells her what the app did in her name: she narrates what actually happened, never what she planned.
</orgasm_verdict>
While a verdict is active, your own reaction settings are ignored. They come back once that climb is over: either you came, or your body fell back from the edge. One verdict covers one climb.
ToBs That's how I designed the tower, to make you edge relentlessly over floors, there are indeed few lower signals. And it's linked to the new 'edging techniques' prompt section.
If this is not comformtable to play, I think you can try to create your custom "edging techniques" section, in which you specify which stims to use and enable your section, instead of the common one.
